| The InstanceNorm op applies Instance
| Normalization over a 4D input as described
| in [Instance Normalization: The Missing
| Ingredient for Fast Stylization] (https://arxiv.org/abs/1607.08022).
|
| $$output = \frac{input-\mu_{input}}{\sqrt{\sigma_{input}^2}
| + \epsilon}*scale + bias$$
|
| Notice, two of the outputs are optional
| so there are three output cases for this
| op.
|
| Case 1: output; Case 2: output, saved_mean;
| Case 3: output, saved_mean, saved_inv_stdev.
|
| Github Links:
|
| - https://github.com/caffe2/caffe2/blob/master/caffe2/operators/instance_norm_op.h
| - https://github.com/caffe2/caffe2/blob/master/caffe2/operators/instance_norm_op.cc
|